Chetan, Madhurima R. http://orcid.org/0000-0002-6868-4969
Dowson, Nicholas
Price, Noah Waterfield
Ather, Sarim
Nicolson, Angus
Gleeson, Fergus V.
Article History
Received: 25 June 2021
Revised: 17 January 2022
Accepted: 2 February 2022
First Online: 3 March 2022
Declarations
:
: The scientific guarantor of this publication is Professor Fergus V Gleeson.
: The authors of this manuscript declare relationships with Optellum Ltd, who are the developers of Lung Cancer Prediction convolutional neural network (LCP-CNN). ND, NWP, and AN are employed by Optellum and FVG has shares in Optellum.Conflicts of interest were mitigated by utilising several publicly available tools developed by others, including the Brock University model, published by a third party and endorsed by the British Thoracic Society guidelines, and the National Lung Screening Trial (NLST) dataset. These were used for the primary analysis within this paper. LCP-CNN is part of a product developed by Optellum that is currently undergoing regulatory review and is expected to become commercially available shortly. LCP-CNN was made available to the group as part of a mutually agreed research collaboration for the secondary analysis within this paper. The raw data and LCP-CNN results are also held by the lead author MC who is an employee at Oxford University Hospital with no links to Optellum.
: One of the authors (ND) has significant statistical expertise.
: Written informed consent was not required for this study because this was an anonymised and retrospective study using the publicly available NLST database.
: Institutional review board approval was not required because this was an anonymised and retrospective study using the publicly available NLST database.
: Some study subjects or cohorts have been previously reported in brief in two conference proceedings at Radiological Society of North America (RSNA) and International Association for the Study of Lung Cancer (IASLC). Both conference proceedings have been included in full within this submission marked as ‘Not for Review’.The conference proceedings reported preliminary work that focused on the adaptation of the U-net deep-learning segmentation algorithm for the automated calculation of diameter using the NLST dataset. The current study significantly expands on this. The current study evaluates the predictive accuracy of the model Brock using automated and manual diameter in comparison to an AI algorithm that does not require diameter measurement, and explores the predictive factors used by the AI model using feature removal in a manner analogous to the covariates used within the Brock model. In addition, the current work studies a more clinically relevant population, by excluding participants with nodules < 6 mm, which do not warrant follow-up, and masses > 30 mm. The authors believe that the greater scope of the current study and its findings are of interest to the readers of <i>European Radiology</i>.We declare that this manuscript is original, has not been published before, and is not currently being considered for publication elsewhere.
: • retrospective• diagnostic or prognostic study• performed at one institution